Fetching the paper…

S$^{3}$: Increasing GPU Utilization during Generative Inference for Higher Throughput · Around